[unreadable] Systematic knowledge of the subcellular locations of proteins and how these locations change in response to drugs, during the cell cycle, during development and during disease will be essential to current comprehensive proteomics efforts. The new field dealing with this subject, Location Proteomics, requires methods for fluorescently-tagging large numbers of proteins, rapidly collecting high-resolution fluorescence microscope images, and, critically, automatically analyzing the resulting distributions. The Murphy group has previously described automated systems that can recognize all major subcellular patterns in 2D and 3D images and can distinguish protein patterns better than visual examination. A critical component of these systems is sets of numerical features that describe each subcellular pattern without being overly sensitive to variations in cell size and shape. The first component of this proposal is research to develop and test solutions to new analysis problems. Within this goal, the first aim is the building of an optimal hierarchical grouping of proteins such that proteins whose patterns are statistically indistinguishable under all conditions are placed in the same group. The second is describing and classifying images based not only on their static patterns but on how those patterns change over time. The third is implementing fast methods for retrieving images from distributed databases based on similarity of protein location patterns. The second component is conversion of current "research-grade" software to "distribution-grade" so that it can be made available to researchers both as stand-alone applications and as part of an image database system. These include pattern classifiers and comparators for image sets. The project team includes leading experts in multimedia database retrieval and data mining and software engineering principles. The software to be developed will be useful not only for large-scale proteomics efforts but also for automated interpretation of traditional cell biology experiments. The hierarchical grouping of proteins by location, in combination with their sequences, will for the first time allow programs for predicting location to operate at the high resolution necessary for discovery of new targeting motifs. The comprehensive approach to location proteomics enabled by this project promises to identify protein differences associated with disease, which can be used as targets for therapies or as markers for early detection and classification of abnormalities. [unreadable] [unreadable]